4 * vfs operations that deal with files
6 * Copyright (C) International Business Machines Corp., 2002,2010
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
45 static inline int cifs_convert_flags(unsigned int flags)
47 if ((flags & O_ACCMODE) == O_RDONLY)
49 else if ((flags & O_ACCMODE) == O_WRONLY)
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
63 static u32 cifs_posix_convert_flags(unsigned int flags)
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
75 posix_flags |= SMB_O_CREAT;
77 posix_flags |= SMB_O_EXCL;
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
88 posix_flags |= SMB_O_DIRECT;
93 static inline int cifs_get_disposition(unsigned int flags)
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
107 static inline int cifs_open_inode_helper(struct inode *inode,
108 struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
109 char *full_path, int xid)
111 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
112 struct timespec temp;
115 if (pCifsInode->clientCanCacheRead) {
116 /* we have the inode open somewhere else
117 no need to discard cache data */
118 goto client_can_cache;
121 /* BB need same check in cifs_create too? */
122 /* if not oplocked, invalidate inode pages if mtime or file
124 temp = cifs_NTtimeToUnix(buf->LastWriteTime);
125 if (timespec_equal(&inode->i_mtime, &temp) &&
127 (loff_t)le64_to_cpu(buf->EndOfFile))) {
128 cFYI(1, "inode unchanged on server");
130 if (inode->i_mapping) {
131 /* BB no need to lock inode until after invalidate
132 since namei code should already have it locked? */
133 rc = filemap_write_and_wait(inode->i_mapping);
135 pCifsInode->write_behind_rc = rc;
137 cFYI(1, "invalidating remote inode since open detected it "
139 invalidate_remote_inode(inode);
144 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
147 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
150 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
151 pCifsInode->clientCanCacheAll = true;
152 pCifsInode->clientCanCacheRead = true;
153 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
154 } else if ((oplock & 0xF) == OPLOCK_READ)
155 pCifsInode->clientCanCacheRead = true;
160 int cifs_posix_open(char *full_path, struct inode **pinode,
161 struct super_block *sb, int mode, unsigned int f_flags,
162 __u32 *poplock, __u16 *pnetfid, int xid)
165 FILE_UNIX_BASIC_INFO *presp_data;
166 __u32 posix_flags = 0;
167 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
168 struct cifs_fattr fattr;
169 struct tcon_link *tlink;
170 struct cifsTconInfo *tcon;
172 cFYI(1, "posix open %s", full_path);
174 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
175 if (presp_data == NULL)
178 tlink = cifs_sb_tlink(cifs_sb);
184 tcon = tlink_tcon(tlink);
185 mode &= ~current_umask();
187 posix_flags = cifs_posix_convert_flags(f_flags);
188 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
189 poplock, full_path, cifs_sb->local_nls,
190 cifs_sb->mnt_cifs_flags &
191 CIFS_MOUNT_MAP_SPECIAL_CHR);
192 cifs_put_tlink(tlink);
197 if (presp_data->Type == cpu_to_le32(-1))
198 goto posix_open_ret; /* open ok, caller does qpathinfo */
201 goto posix_open_ret; /* caller does not need info */
203 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
205 /* get new inode and set it up */
206 if (*pinode == NULL) {
207 cifs_fill_uniqueid(sb, &fattr);
208 *pinode = cifs_iget(sb, &fattr);
214 cifs_fattr_to_inode(*pinode, &fattr);
222 struct cifsFileInfo *
223 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
224 struct tcon_link *tlink, __u32 oplock)
226 struct dentry *dentry = file->f_path.dentry;
227 struct inode *inode = dentry->d_inode;
228 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
229 struct cifsFileInfo *pCifsFile;
231 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
232 if (pCifsFile == NULL)
235 pCifsFile->netfid = fileHandle;
236 pCifsFile->pid = current->tgid;
237 pCifsFile->uid = current_fsuid();
238 pCifsFile->dentry = dget(dentry);
239 pCifsFile->f_flags = file->f_flags;
240 pCifsFile->invalidHandle = false;
241 pCifsFile->closePend = false;
242 pCifsFile->tlink = cifs_get_tlink(tlink);
243 mutex_init(&pCifsFile->fh_mutex);
244 mutex_init(&pCifsFile->lock_mutex);
245 INIT_LIST_HEAD(&pCifsFile->llist);
246 atomic_set(&pCifsFile->count, 1);
247 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
249 spin_lock(&cifs_file_list_lock);
250 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
251 /* if readable file instance put first in list*/
252 if (file->f_mode & FMODE_READ)
253 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
255 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
256 spin_unlock(&cifs_file_list_lock);
258 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
259 pCifsInode->clientCanCacheAll = true;
260 pCifsInode->clientCanCacheRead = true;
261 cFYI(1, "Exclusive Oplock inode %p", inode);
262 } else if ((oplock & 0xF) == OPLOCK_READ)
263 pCifsInode->clientCanCacheRead = true;
265 file->private_data = pCifsFile;
269 /* Release a reference on the file private data */
270 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
272 if (atomic_dec_and_test(&cifs_file->count)) {
273 cifs_put_tlink(cifs_file->tlink);
274 dput(cifs_file->dentry);
279 int cifs_open(struct inode *inode, struct file *file)
284 struct cifs_sb_info *cifs_sb;
285 struct cifsTconInfo *tcon;
286 struct tcon_link *tlink;
287 struct cifsFileInfo *pCifsFile = NULL;
288 struct cifsInodeInfo *pCifsInode;
289 char *full_path = NULL;
293 FILE_ALL_INFO *buf = NULL;
297 cifs_sb = CIFS_SB(inode->i_sb);
298 tlink = cifs_sb_tlink(cifs_sb);
301 return PTR_ERR(tlink);
303 tcon = tlink_tcon(tlink);
305 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
307 full_path = build_path_from_dentry(file->f_path.dentry);
308 if (full_path == NULL) {
313 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
314 inode, file->f_flags, full_path);
321 if (!tcon->broken_posix_open && tcon->unix_ext &&
322 (tcon->ses->capabilities & CAP_UNIX) &&
323 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
324 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
325 /* can not refresh inode info since size could be stale */
326 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
327 cifs_sb->mnt_file_mode /* ignored */,
328 file->f_flags, &oplock, &netfid, xid);
330 cFYI(1, "posix open succeeded");
332 pCifsFile = cifs_new_fileinfo(netfid, file, tlink,
334 if (pCifsFile == NULL) {
335 CIFSSMBClose(xid, tcon, netfid);
339 cifs_fscache_set_inode_cookie(inode, file);
342 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
343 if (tcon->ses->serverNOS)
344 cERROR(1, "server %s of type %s returned"
345 " unexpected error on SMB posix open"
346 ", disabling posix open support."
347 " Check if server update available.",
348 tcon->ses->serverName,
349 tcon->ses->serverNOS);
350 tcon->broken_posix_open = true;
351 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
352 (rc != -EOPNOTSUPP)) /* path not found or net err */
354 /* else fallthrough to retry open the old way on network i/o
358 desiredAccess = cifs_convert_flags(file->f_flags);
360 /*********************************************************************
361 * open flag mapping table:
363 * POSIX Flag CIFS Disposition
364 * ---------- ----------------
365 * O_CREAT FILE_OPEN_IF
366 * O_CREAT | O_EXCL FILE_CREATE
367 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
368 * O_TRUNC FILE_OVERWRITE
369 * none of the above FILE_OPEN
371 * Note that there is not a direct match between disposition
372 * FILE_SUPERSEDE (ie create whether or not file exists although
373 * O_CREAT | O_TRUNC is similar but truncates the existing
374 * file rather than creating a new file as FILE_SUPERSEDE does
375 * (which uses the attributes / metadata passed in on open call)
377 *? O_SYNC is a reasonable match to CIFS writethrough flag
378 *? and the read write flags match reasonably. O_LARGEFILE
379 *? is irrelevant because largefile support is always used
380 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
381 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
382 *********************************************************************/
384 disposition = cifs_get_disposition(file->f_flags);
386 /* BB pass O_SYNC flag through on file attributes .. BB */
388 /* Also refresh inode by passing in file_info buf returned by SMBOpen
389 and calling get_inode_info with returned buf (at least helps
390 non-Unix server case) */
392 /* BB we can not do this if this is the second open of a file
393 and the first handle has writebehind data, we might be
394 able to simply do a filemap_fdatawrite/filemap_fdatawait first */
395 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
401 if (tcon->ses->capabilities & CAP_NT_SMBS)
402 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
403 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
404 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
405 & CIFS_MOUNT_MAP_SPECIAL_CHR);
407 rc = -EIO; /* no NT SMB support fall into legacy open below */
410 /* Old server, try legacy style OpenX */
411 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
412 desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
413 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
414 & CIFS_MOUNT_MAP_SPECIAL_CHR);
417 cFYI(1, "cifs_open returned 0x%x", rc);
421 rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
425 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
426 if (pCifsFile == NULL) {
431 cifs_fscache_set_inode_cookie(inode, file);
433 if (oplock & CIFS_CREATE_ACTION) {
434 /* time to set mode which we can not set earlier due to
435 problems creating new read-only files */
436 if (tcon->unix_ext) {
437 struct cifs_unix_set_info_args args = {
438 .mode = inode->i_mode,
441 .ctime = NO_CHANGE_64,
442 .atime = NO_CHANGE_64,
443 .mtime = NO_CHANGE_64,
446 CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
448 cifs_sb->mnt_cifs_flags &
449 CIFS_MOUNT_MAP_SPECIAL_CHR);
457 cifs_put_tlink(tlink);
461 /* Try to reacquire byte range locks that were released when session */
462 /* to server was lost */
463 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
467 /* BB list all locks open on this file and relock */
472 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
477 struct cifs_sb_info *cifs_sb;
478 struct cifsTconInfo *tcon;
479 struct cifsInodeInfo *pCifsInode;
481 char *full_path = NULL;
483 int disposition = FILE_OPEN;
487 mutex_lock(&pCifsFile->fh_mutex);
488 if (!pCifsFile->invalidHandle) {
489 mutex_unlock(&pCifsFile->fh_mutex);
495 inode = pCifsFile->dentry->d_inode;
496 cifs_sb = CIFS_SB(inode->i_sb);
497 tcon = tlink_tcon(pCifsFile->tlink);
499 /* can not grab rename sem here because various ops, including
500 those that already have the rename sem can end up causing writepage
501 to get called and if the server was down that means we end up here,
502 and we can never tell if the caller already has the rename_sem */
503 full_path = build_path_from_dentry(pCifsFile->dentry);
504 if (full_path == NULL) {
506 mutex_unlock(&pCifsFile->fh_mutex);
511 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
512 inode, pCifsFile->f_flags, full_path);
519 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
520 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
521 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
524 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
525 * original open. Must mask them off for a reopen.
527 unsigned int oflags = pCifsFile->f_flags &
528 ~(O_CREAT | O_EXCL | O_TRUNC);
530 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
531 cifs_sb->mnt_file_mode /* ignored */,
532 oflags, &oplock, &netfid, xid);
534 cFYI(1, "posix reopen succeeded");
537 /* fallthrough to retry open the old way on errors, especially
538 in the reconnect path it is important to retry hard */
541 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
543 /* Can not refresh inode by passing in file_info buf to be returned
544 by SMBOpen and then calling get_inode_info with returned buf
545 since file might have write behind data that needs to be flushed
546 and server version of file size can be stale. If we knew for sure
547 that inode was not dirty locally we could do this */
549 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
550 CREATE_NOT_DIR, &netfid, &oplock, NULL,
551 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
552 CIFS_MOUNT_MAP_SPECIAL_CHR);
554 mutex_unlock(&pCifsFile->fh_mutex);
555 cFYI(1, "cifs_open returned 0x%x", rc);
556 cFYI(1, "oplock: %d", oplock);
557 goto reopen_error_exit;
561 pCifsFile->netfid = netfid;
562 pCifsFile->invalidHandle = false;
563 mutex_unlock(&pCifsFile->fh_mutex);
564 pCifsInode = CIFS_I(inode);
567 rc = filemap_write_and_wait(inode->i_mapping);
569 CIFS_I(inode)->write_behind_rc = rc;
571 pCifsInode->clientCanCacheAll = false;
572 pCifsInode->clientCanCacheRead = false;
574 rc = cifs_get_inode_info_unix(&inode,
575 full_path, inode->i_sb, xid);
577 rc = cifs_get_inode_info(&inode,
578 full_path, NULL, inode->i_sb,
580 } /* else we are writing out data to server already
581 and could deadlock if we tried to flush data, and
582 since we do not know if we have data that would
583 invalidate the current end of file on the server
584 we can not go to the server to get the new inod
586 if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
587 pCifsInode->clientCanCacheAll = true;
588 pCifsInode->clientCanCacheRead = true;
589 cFYI(1, "Exclusive Oplock granted on inode %p",
590 pCifsFile->dentry->d_inode);
591 } else if ((oplock & 0xF) == OPLOCK_READ) {
592 pCifsInode->clientCanCacheRead = true;
593 pCifsInode->clientCanCacheAll = false;
595 pCifsInode->clientCanCacheRead = false;
596 pCifsInode->clientCanCacheAll = false;
598 cifs_relock_file(pCifsFile);
606 int cifs_close(struct inode *inode, struct file *file)
610 struct cifs_sb_info *cifs_sb;
611 struct cifsTconInfo *pTcon;
612 struct cifsFileInfo *pSMBFile = file->private_data;
616 cifs_sb = CIFS_SB(inode->i_sb);
617 pTcon = tlink_tcon(pSMBFile->tlink);
619 struct cifsLockInfo *li, *tmp;
620 spin_lock(&cifs_file_list_lock);
621 pSMBFile->closePend = true;
623 /* no sense reconnecting to close a file that is
625 if (!pTcon->need_reconnect) {
626 spin_unlock(&cifs_file_list_lock);
628 while ((atomic_read(&pSMBFile->count) != 1)
629 && (timeout <= 2048)) {
630 /* Give write a better chance to get to
631 server ahead of the close. We do not
632 want to add a wait_q here as it would
633 increase the memory utilization as
634 the struct would be in each open file,
635 but this should give enough time to
637 cFYI(DBG2, "close delay, write pending");
641 if (!pTcon->need_reconnect &&
642 !pSMBFile->invalidHandle)
643 rc = CIFSSMBClose(xid, pTcon,
646 spin_unlock(&cifs_file_list_lock);
648 spin_unlock(&cifs_file_list_lock);
650 /* Delete any outstanding lock records.
651 We'll lose them when the file is closed anyway. */
652 mutex_lock(&pSMBFile->lock_mutex);
653 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
654 list_del(&li->llist);
657 mutex_unlock(&pSMBFile->lock_mutex);
659 spin_lock(&cifs_file_list_lock);
660 list_del(&pSMBFile->flist);
661 list_del(&pSMBFile->tlist);
662 spin_unlock(&cifs_file_list_lock);
663 cifsFileInfo_put(file->private_data);
664 file->private_data = NULL;
668 spin_lock(&cifs_file_list_lock);
669 if (list_empty(&(CIFS_I(inode)->openFileList))) {
670 cFYI(1, "closing last open instance for inode %p", inode);
671 /* if the file is not open we do not know if we can cache info
672 on this inode, much less write behind and read ahead */
673 CIFS_I(inode)->clientCanCacheRead = false;
674 CIFS_I(inode)->clientCanCacheAll = false;
676 spin_unlock(&cifs_file_list_lock);
677 if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
678 rc = CIFS_I(inode)->write_behind_rc;
683 int cifs_closedir(struct inode *inode, struct file *file)
687 struct cifsFileInfo *pCFileStruct = file->private_data;
690 cFYI(1, "Closedir inode = 0x%p", inode);
695 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
697 cFYI(1, "Freeing private data in close dir");
698 spin_lock(&cifs_file_list_lock);
699 if (!pCFileStruct->srch_inf.endOfSearch &&
700 !pCFileStruct->invalidHandle) {
701 pCFileStruct->invalidHandle = true;
702 spin_unlock(&cifs_file_list_lock);
703 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
704 cFYI(1, "Closing uncompleted readdir with rc %d",
706 /* not much we can do if it fails anyway, ignore rc */
709 spin_unlock(&cifs_file_list_lock);
710 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
712 cFYI(1, "closedir free smb buf in srch struct");
713 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
714 if (pCFileStruct->srch_inf.smallBuf)
715 cifs_small_buf_release(ptmp);
717 cifs_buf_release(ptmp);
719 cifs_put_tlink(pCFileStruct->tlink);
720 kfree(file->private_data);
721 file->private_data = NULL;
723 /* BB can we lock the filestruct while this is going on? */
728 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
729 __u64 offset, __u8 lockType)
731 struct cifsLockInfo *li =
732 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
738 mutex_lock(&fid->lock_mutex);
739 list_add(&li->llist, &fid->llist);
740 mutex_unlock(&fid->lock_mutex);
744 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
750 bool wait_flag = false;
751 struct cifs_sb_info *cifs_sb;
752 struct cifsTconInfo *tcon;
754 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
755 bool posix_locking = 0;
757 length = 1 + pfLock->fl_end - pfLock->fl_start;
761 cFYI(1, "Lock parm: 0x%x flockflags: "
762 "0x%x flocktype: 0x%x start: %lld end: %lld",
763 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
766 if (pfLock->fl_flags & FL_POSIX)
768 if (pfLock->fl_flags & FL_FLOCK)
770 if (pfLock->fl_flags & FL_SLEEP) {
771 cFYI(1, "Blocking lock");
774 if (pfLock->fl_flags & FL_ACCESS)
775 cFYI(1, "Process suspended by mandatory locking - "
776 "not implemented yet");
777 if (pfLock->fl_flags & FL_LEASE)
778 cFYI(1, "Lease on file - not implemented yet");
779 if (pfLock->fl_flags &
780 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
781 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
783 if (pfLock->fl_type == F_WRLCK) {
786 } else if (pfLock->fl_type == F_UNLCK) {
789 /* Check if unlock includes more than
791 } else if (pfLock->fl_type == F_RDLCK) {
793 lockType |= LOCKING_ANDX_SHARED_LOCK;
795 } else if (pfLock->fl_type == F_EXLCK) {
798 } else if (pfLock->fl_type == F_SHLCK) {
800 lockType |= LOCKING_ANDX_SHARED_LOCK;
803 cFYI(1, "Unknown type of lock");
805 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
806 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
808 if (file->private_data == NULL) {
813 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
815 if ((tcon->ses->capabilities & CAP_UNIX) &&
816 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
817 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
819 /* BB add code here to normalize offset and length to
820 account for negative length which we can not accept over the
825 if (lockType & LOCKING_ANDX_SHARED_LOCK)
826 posix_lock_type = CIFS_RDLCK;
828 posix_lock_type = CIFS_WRLCK;
829 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
831 posix_lock_type, wait_flag);
836 /* BB we could chain these into one lock request BB */
837 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
838 0, 1, lockType, 0 /* wait flag */ );
840 rc = CIFSSMBLock(xid, tcon, netfid, length,
841 pfLock->fl_start, 1 /* numUnlock */ ,
842 0 /* numLock */ , lockType,
844 pfLock->fl_type = F_UNLCK;
846 cERROR(1, "Error unlocking previously locked "
847 "range %d during test of lock", rc);
851 /* if rc == ERR_SHARING_VIOLATION ? */
854 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
855 pfLock->fl_type = F_WRLCK;
857 rc = CIFSSMBLock(xid, tcon, netfid, length,
858 pfLock->fl_start, 0, 1,
859 lockType | LOCKING_ANDX_SHARED_LOCK,
862 rc = CIFSSMBLock(xid, tcon, netfid,
863 length, pfLock->fl_start, 1, 0,
865 LOCKING_ANDX_SHARED_LOCK,
867 pfLock->fl_type = F_RDLCK;
869 cERROR(1, "Error unlocking "
870 "previously locked range %d "
871 "during test of lock", rc);
874 pfLock->fl_type = F_WRLCK;
884 if (!numLock && !numUnlock) {
885 /* if no lock or unlock then nothing
886 to do since we do not know what it is */
893 if (lockType & LOCKING_ANDX_SHARED_LOCK)
894 posix_lock_type = CIFS_RDLCK;
896 posix_lock_type = CIFS_WRLCK;
899 posix_lock_type = CIFS_UNLCK;
901 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
903 posix_lock_type, wait_flag);
905 struct cifsFileInfo *fid = file->private_data;
908 rc = CIFSSMBLock(xid, tcon, netfid, length,
910 0, numLock, lockType, wait_flag);
913 /* For Windows locks we must store them. */
914 rc = store_file_lock(fid, length,
915 pfLock->fl_start, lockType);
917 } else if (numUnlock) {
918 /* For each stored lock that this unlock overlaps
919 completely, unlock it. */
921 struct cifsLockInfo *li, *tmp;
924 mutex_lock(&fid->lock_mutex);
925 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
926 if (pfLock->fl_start <= li->offset &&
927 (pfLock->fl_start + length) >=
928 (li->offset + li->length)) {
929 stored_rc = CIFSSMBLock(xid, tcon,
931 li->length, li->offset,
932 1, 0, li->type, false);
936 list_del(&li->llist);
941 mutex_unlock(&fid->lock_mutex);
945 if (pfLock->fl_flags & FL_POSIX)
946 posix_lock_file_wait(file, pfLock);
952 * Set the timeout on write requests past EOF. For some servers (Windows)
953 * these calls can be very long.
955 * If we're writing >10M past the EOF we give a 180s timeout. Anything less
956 * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
957 * The 10M cutoff is totally arbitrary. A better scheme for this would be
958 * welcome if someone wants to suggest one.
960 * We may be able to do a better job with this if there were some way to
961 * declare that a file should be sparse.
964 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
966 if (offset <= cifsi->server_eof)
968 else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
969 return CIFS_VLONG_OP;
974 /* update the file size (if needed) after a write */
976 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
977 unsigned int bytes_written)
979 loff_t end_of_write = offset + bytes_written;
981 if (end_of_write > cifsi->server_eof)
982 cifsi->server_eof = end_of_write;
985 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
986 size_t write_size, loff_t *poffset)
989 unsigned int bytes_written = 0;
990 unsigned int total_written;
991 struct cifs_sb_info *cifs_sb;
992 struct cifsTconInfo *pTcon;
994 struct cifsFileInfo *open_file;
995 struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
997 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
999 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
1000 *poffset, file->f_path.dentry->d_name.name); */
1002 if (file->private_data == NULL)
1005 open_file = file->private_data;
1006 pTcon = tlink_tcon(open_file->tlink);
1008 rc = generic_write_checks(file, poffset, &write_size, 0);
1014 long_op = cifs_write_timeout(cifsi, *poffset);
1015 for (total_written = 0; write_size > total_written;
1016 total_written += bytes_written) {
1018 while (rc == -EAGAIN) {
1019 if (file->private_data == NULL) {
1020 /* file has been closed on us */
1022 /* if we have gotten here we have written some data
1023 and blocked, and the file has been freed on us while
1024 we blocked so return what we managed to write */
1025 return total_written;
1027 if (open_file->closePend) {
1030 return total_written;
1034 if (open_file->invalidHandle) {
1035 /* we could deadlock if we called
1036 filemap_fdatawait from here so tell
1037 reopen_file not to flush data to server
1039 rc = cifs_reopen_file(open_file, false);
1044 rc = CIFSSMBWrite(xid, pTcon,
1046 min_t(const int, cifs_sb->wsize,
1047 write_size - total_written),
1048 *poffset, &bytes_written,
1049 NULL, write_data + total_written, long_op);
1051 if (rc || (bytes_written == 0)) {
1059 cifs_update_eof(cifsi, *poffset, bytes_written);
1060 *poffset += bytes_written;
1062 long_op = CIFS_STD_OP; /* subsequent writes fast -
1063 15 seconds is plenty */
1066 cifs_stats_bytes_written(pTcon, total_written);
1068 /* since the write may have blocked check these pointers again */
1069 if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1070 struct inode *inode = file->f_path.dentry->d_inode;
1071 /* Do not update local mtime - server will set its actual value on write
1072 * inode->i_ctime = inode->i_mtime =
1073 * current_fs_time(inode->i_sb);*/
1074 if (total_written > 0) {
1075 spin_lock(&inode->i_lock);
1076 if (*poffset > file->f_path.dentry->d_inode->i_size)
1077 i_size_write(file->f_path.dentry->d_inode,
1079 spin_unlock(&inode->i_lock);
1081 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1084 return total_written;
1087 static ssize_t cifs_write(struct cifsFileInfo *open_file,
1088 const char *write_data, size_t write_size,
1092 unsigned int bytes_written = 0;
1093 unsigned int total_written;
1094 struct cifs_sb_info *cifs_sb;
1095 struct cifsTconInfo *pTcon;
1097 struct dentry *dentry = open_file->dentry;
1098 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1100 cifs_sb = CIFS_SB(dentry->d_sb);
1102 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1103 *poffset, dentry->d_name.name);
1105 pTcon = tlink_tcon(open_file->tlink);
1109 long_op = cifs_write_timeout(cifsi, *poffset);
1110 for (total_written = 0; write_size > total_written;
1111 total_written += bytes_written) {
1113 while (rc == -EAGAIN) {
1114 if (open_file->closePend) {
1117 return total_written;
1121 if (open_file->invalidHandle) {
1122 /* we could deadlock if we called
1123 filemap_fdatawait from here so tell
1124 reopen_file not to flush data to
1126 rc = cifs_reopen_file(open_file, false);
1130 if (experimEnabled || (pTcon->ses->server &&
1131 ((pTcon->ses->server->secMode &
1132 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1137 len = min((size_t)cifs_sb->wsize,
1138 write_size - total_written);
1139 /* iov[0] is reserved for smb header */
1140 iov[1].iov_base = (char *)write_data +
1142 iov[1].iov_len = len;
1143 rc = CIFSSMBWrite2(xid, pTcon,
1144 open_file->netfid, len,
1145 *poffset, &bytes_written,
1148 rc = CIFSSMBWrite(xid, pTcon,
1150 min_t(const int, cifs_sb->wsize,
1151 write_size - total_written),
1152 *poffset, &bytes_written,
1153 write_data + total_written,
1156 if (rc || (bytes_written == 0)) {
1164 cifs_update_eof(cifsi, *poffset, bytes_written);
1165 *poffset += bytes_written;
1167 long_op = CIFS_STD_OP; /* subsequent writes fast -
1168 15 seconds is plenty */
1171 cifs_stats_bytes_written(pTcon, total_written);
1173 if (total_written > 0) {
1174 spin_lock(&dentry->d_inode->i_lock);
1175 if (*poffset > dentry->d_inode->i_size)
1176 i_size_write(dentry->d_inode, *poffset);
1177 spin_unlock(&dentry->d_inode->i_lock);
1179 mark_inode_dirty_sync(dentry->d_inode);
1181 return total_written;
1184 #ifdef CONFIG_CIFS_EXPERIMENTAL
1185 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1188 struct cifsFileInfo *open_file = NULL;
1189 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1191 /* only filter by fsuid on multiuser mounts */
1192 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1195 spin_lock(&cifs_file_list_lock);
1196 /* we could simply get the first_list_entry since write-only entries
1197 are always at the end of the list but since the first entry might
1198 have a close pending, we go through the whole list */
1199 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1200 if (open_file->closePend)
1202 if (fsuid_only && open_file->uid != current_fsuid())
1204 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1205 if (!open_file->invalidHandle) {
1206 /* found a good file */
1207 /* lock it so it will not be closed on us */
1208 cifsFileInfo_get(open_file);
1209 spin_unlock(&cifs_file_list_lock);
1211 } /* else might as well continue, and look for
1212 another, or simply have the caller reopen it
1213 again rather than trying to fix this handle */
1214 } else /* write only file */
1215 break; /* write only files are last so must be done */
1217 spin_unlock(&cifs_file_list_lock);
1222 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1225 struct cifsFileInfo *open_file;
1226 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1227 bool any_available = false;
1230 /* Having a null inode here (because mapping->host was set to zero by
1231 the VFS or MM) should not happen but we had reports of on oops (due to
1232 it being zero) during stress testcases so we need to check for it */
1234 if (cifs_inode == NULL) {
1235 cERROR(1, "Null inode passed to cifs_writeable_file");
1240 /* only filter by fsuid on multiuser mounts */
1241 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1244 spin_lock(&cifs_file_list_lock);
1246 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1247 if (open_file->closePend)
1249 if (!any_available && open_file->pid != current->tgid)
1251 if (fsuid_only && open_file->uid != current_fsuid())
1253 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1254 cifsFileInfo_get(open_file);
1256 if (!open_file->invalidHandle) {
1257 /* found a good writable file */
1258 spin_unlock(&cifs_file_list_lock);
1262 spin_unlock(&cifs_file_list_lock);
1263 /* Had to unlock since following call can block */
1264 rc = cifs_reopen_file(open_file, false);
1266 if (!open_file->closePend)
1268 else { /* start over in case this was deleted */
1269 /* since the list could be modified */
1270 spin_lock(&cifs_file_list_lock);
1271 cifsFileInfo_put(open_file);
1272 goto refind_writable;
1276 /* if it fails, try another handle if possible -
1277 (we can not do this if closePending since
1278 loop could be modified - in which case we
1279 have to start at the beginning of the list
1280 again. Note that it would be bad
1281 to hold up writepages here (rather than
1282 in caller) with continuous retries */
1283 cFYI(1, "wp failed on reopen file");
1284 spin_lock(&cifs_file_list_lock);
1285 /* can not use this handle, no write
1286 pending on this one after all */
1287 cifsFileInfo_put(open_file);
1289 if (open_file->closePend) /* list could have changed */
1290 goto refind_writable;
1291 /* else we simply continue to the next entry. Thus
1292 we do not loop on reopen errors. If we
1293 can not reopen the file, for example if we
1294 reconnected to a server with another client
1295 racing to delete or lock the file we would not
1296 make progress if we restarted before the beginning
1297 of the loop here. */
1300 /* couldn't find useable FH with same pid, try any available */
1301 if (!any_available) {
1302 any_available = true;
1303 goto refind_writable;
1305 spin_unlock(&cifs_file_list_lock);
1309 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1311 struct address_space *mapping = page->mapping;
1312 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1315 int bytes_written = 0;
1316 struct cifs_sb_info *cifs_sb;
1317 struct inode *inode;
1318 struct cifsFileInfo *open_file;
1320 if (!mapping || !mapping->host)
1323 inode = page->mapping->host;
1324 cifs_sb = CIFS_SB(inode->i_sb);
1326 offset += (loff_t)from;
1327 write_data = kmap(page);
1330 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1335 /* racing with truncate? */
1336 if (offset > mapping->host->i_size) {
1338 return 0; /* don't care */
1341 /* check to make sure that we are not extending the file */
1342 if (mapping->host->i_size - offset < (loff_t)to)
1343 to = (unsigned)(mapping->host->i_size - offset);
1345 open_file = find_writable_file(CIFS_I(mapping->host), false);
1347 bytes_written = cifs_write(open_file, write_data,
1348 to - from, &offset);
1349 cifsFileInfo_put(open_file);
1350 /* Does mm or vfs already set times? */
1351 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1352 if ((bytes_written > 0) && (offset))
1354 else if (bytes_written < 0)
1357 cFYI(1, "No writeable filehandles for inode");
1365 static int cifs_writepages(struct address_space *mapping,
1366 struct writeback_control *wbc)
1368 struct backing_dev_info *bdi = mapping->backing_dev_info;
1369 unsigned int bytes_to_write;
1370 unsigned int bytes_written;
1371 struct cifs_sb_info *cifs_sb;
1375 int range_whole = 0;
1382 struct cifsFileInfo *open_file;
1383 struct cifsTconInfo *tcon;
1384 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1386 struct pagevec pvec;
1392 * BB: Is this meaningful for a non-block-device file system?
1393 * If it is, we should test it again after we do I/O
1395 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1396 wbc->encountered_congestion = 1;
1400 cifs_sb = CIFS_SB(mapping->host->i_sb);
1403 * If wsize is smaller that the page cache size, default to writing
1404 * one page at a time via cifs_writepage
1406 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1407 return generic_writepages(mapping, wbc);
1409 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1411 return generic_writepages(mapping, wbc);
1414 * if there's no open file, then this is likely to fail too,
1415 * but it'll at least handle the return. Maybe it should be
1418 open_file = find_writable_file(CIFS_I(mapping->host), false);
1421 return generic_writepages(mapping, wbc);
1424 tcon = tlink_tcon(open_file->tlink);
1425 if (!experimEnabled && tcon->ses->server->secMode &
1426 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1427 cifsFileInfo_put(open_file);
1428 return generic_writepages(mapping, wbc);
1430 cifsFileInfo_put(open_file);
1434 pagevec_init(&pvec, 0);
1435 if (wbc->range_cyclic) {
1436 index = mapping->writeback_index; /* Start from prev offset */
1439 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1440 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1441 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1446 while (!done && (index <= end) &&
1447 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1448 PAGECACHE_TAG_DIRTY,
1449 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1458 for (i = 0; i < nr_pages; i++) {
1459 page = pvec.pages[i];
1461 * At this point we hold neither mapping->tree_lock nor
1462 * lock on the page itself: the page may be truncated or
1463 * invalidated (changing page->mapping to NULL), or even
1464 * swizzled back from swapper_space to tmpfs file
1470 else if (!trylock_page(page))
1473 if (unlikely(page->mapping != mapping)) {
1478 if (!wbc->range_cyclic && page->index > end) {
1484 if (next && (page->index != next)) {
1485 /* Not next consecutive page */
1490 if (wbc->sync_mode != WB_SYNC_NONE)
1491 wait_on_page_writeback(page);
1493 if (PageWriteback(page) ||
1494 !clear_page_dirty_for_io(page)) {
1500 * This actually clears the dirty bit in the radix tree.
1501 * See cifs_writepage() for more commentary.
1503 set_page_writeback(page);
1505 if (page_offset(page) >= mapping->host->i_size) {
1508 end_page_writeback(page);
1513 * BB can we get rid of this? pages are held by pvec
1515 page_cache_get(page);
1517 len = min(mapping->host->i_size - page_offset(page),
1518 (loff_t)PAGE_CACHE_SIZE);
1520 /* reserve iov[0] for the smb header */
1522 iov[n_iov].iov_base = kmap(page);
1523 iov[n_iov].iov_len = len;
1524 bytes_to_write += len;
1528 offset = page_offset(page);
1530 next = page->index + 1;
1531 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1535 open_file = find_writable_file(CIFS_I(mapping->host),
1538 cERROR(1, "No writable handles for inode");
1541 long_op = cifs_write_timeout(cifsi, offset);
1542 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1543 bytes_to_write, offset,
1544 &bytes_written, iov, n_iov,
1546 cifsFileInfo_put(open_file);
1547 cifs_update_eof(cifsi, offset, bytes_written);
1550 if (rc || bytes_written < bytes_to_write) {
1551 cERROR(1, "Write2 ret %d, wrote %d",
1553 /* BB what if continued retry is
1554 requested via mount flags? */
1556 set_bit(AS_ENOSPC, &mapping->flags);
1558 set_bit(AS_EIO, &mapping->flags);
1560 cifs_stats_bytes_written(tcon, bytes_written);
1563 for (i = 0; i < n_iov; i++) {
1564 page = pvec.pages[first + i];
1565 /* Should we also set page error on
1566 success rc but too little data written? */
1567 /* BB investigate retry logic on temporary
1568 server crash cases and how recovery works
1569 when page marked as error */
1574 end_page_writeback(page);
1575 page_cache_release(page);
1577 if ((wbc->nr_to_write -= n_iov) <= 0)
1581 /* Need to re-find the pages we skipped */
1582 index = pvec.pages[0]->index + 1;
1584 pagevec_release(&pvec);
1586 if (!scanned && !done) {
1588 * We hit the last page and there is more work to be done: wrap
1589 * back to the start of the file
1595 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1596 mapping->writeback_index = index;
1603 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1609 /* BB add check for wbc flags */
1610 page_cache_get(page);
1611 if (!PageUptodate(page))
1612 cFYI(1, "ppw - page not up to date");
1615 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1617 * A writepage() implementation always needs to do either this,
1618 * or re-dirty the page with "redirty_page_for_writepage()" in
1619 * the case of a failure.
1621 * Just unlocking the page will cause the radix tree tag-bits
1622 * to fail to update with the state of the page correctly.
1624 set_page_writeback(page);
1625 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1626 SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1628 end_page_writeback(page);
1629 page_cache_release(page);
1634 static int cifs_write_end(struct file *file, struct address_space *mapping,
1635 loff_t pos, unsigned len, unsigned copied,
1636 struct page *page, void *fsdata)
1639 struct inode *inode = mapping->host;
1641 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1644 if (PageChecked(page)) {
1646 SetPageUptodate(page);
1647 ClearPageChecked(page);
1648 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1649 SetPageUptodate(page);
1651 if (!PageUptodate(page)) {
1653 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1657 /* this is probably better than directly calling
1658 partialpage_write since in this function the file handle is
1659 known which we might as well leverage */
1660 /* BB check if anything else missing out of ppw
1661 such as updating last write time */
1662 page_data = kmap(page);
1663 rc = cifs_write(file->private_data, page_data + offset,
1665 /* if (rc < 0) should we set writebehind rc? */
1672 set_page_dirty(page);
1676 spin_lock(&inode->i_lock);
1677 if (pos > inode->i_size)
1678 i_size_write(inode, pos);
1679 spin_unlock(&inode->i_lock);
1683 page_cache_release(page);
1688 int cifs_fsync(struct file *file, int datasync)
1692 struct cifsTconInfo *tcon;
1693 struct cifsFileInfo *smbfile = file->private_data;
1694 struct inode *inode = file->f_path.dentry->d_inode;
1698 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1699 file->f_path.dentry->d_name.name, datasync);
1701 rc = filemap_write_and_wait(inode->i_mapping);
1703 rc = CIFS_I(inode)->write_behind_rc;
1704 CIFS_I(inode)->write_behind_rc = 0;
1705 tcon = tlink_tcon(smbfile->tlink);
1706 if (!rc && tcon && smbfile &&
1707 !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1708 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1715 /* static void cifs_sync_page(struct page *page)
1717 struct address_space *mapping;
1718 struct inode *inode;
1719 unsigned long index = page->index;
1720 unsigned int rpages = 0;
1723 cFYI(1, "sync page %p", page);
1724 mapping = page->mapping;
1727 inode = mapping->host;
1731 /* fill in rpages then
1732 result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1734 /* cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1744 * As file closes, flush all cached write data for this inode checking
1745 * for write behind errors.
1747 int cifs_flush(struct file *file, fl_owner_t id)
1749 struct inode *inode = file->f_path.dentry->d_inode;
1752 /* Rather than do the steps manually:
1753 lock the inode for writing
1754 loop through pages looking for write behind data (dirty pages)
1755 coalesce into contiguous 16K (or smaller) chunks to write to server
1756 send to server (prefer in parallel)
1757 deal with writebehind errors
1758 unlock inode for writing
1759 filemapfdatawrite appears easier for the time being */
1761 rc = filemap_fdatawrite(inode->i_mapping);
1762 /* reset wb rc if we were able to write out dirty pages */
1764 rc = CIFS_I(inode)->write_behind_rc;
1765 CIFS_I(inode)->write_behind_rc = 0;
1768 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1773 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1774 size_t read_size, loff_t *poffset)
1777 unsigned int bytes_read = 0;
1778 unsigned int total_read = 0;
1779 unsigned int current_read_size;
1780 struct cifs_sb_info *cifs_sb;
1781 struct cifsTconInfo *pTcon;
1783 struct cifsFileInfo *open_file;
1784 char *smb_read_data;
1785 char __user *current_offset;
1786 struct smb_com_read_rsp *pSMBr;
1789 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1791 if (file->private_data == NULL) {
1796 open_file = file->private_data;
1797 pTcon = tlink_tcon(open_file->tlink);
1799 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1800 cFYI(1, "attempting read on write only file instance");
1802 for (total_read = 0, current_offset = read_data;
1803 read_size > total_read;
1804 total_read += bytes_read, current_offset += bytes_read) {
1805 current_read_size = min_t(const int, read_size - total_read,
1808 smb_read_data = NULL;
1809 while (rc == -EAGAIN) {
1810 int buf_type = CIFS_NO_BUFFER;
1811 if ((open_file->invalidHandle) &&
1812 (!open_file->closePend)) {
1813 rc = cifs_reopen_file(open_file, true);
1817 rc = CIFSSMBRead(xid, pTcon,
1819 current_read_size, *poffset,
1820 &bytes_read, &smb_read_data,
1822 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1823 if (smb_read_data) {
1824 if (copy_to_user(current_offset,
1826 4 /* RFC1001 length field */ +
1827 le16_to_cpu(pSMBr->DataOffset),
1831 if (buf_type == CIFS_SMALL_BUFFER)
1832 cifs_small_buf_release(smb_read_data);
1833 else if (buf_type == CIFS_LARGE_BUFFER)
1834 cifs_buf_release(smb_read_data);
1835 smb_read_data = NULL;
1838 if (rc || (bytes_read == 0)) {
1846 cifs_stats_bytes_read(pTcon, bytes_read);
1847 *poffset += bytes_read;
1855 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1859 unsigned int bytes_read = 0;
1860 unsigned int total_read;
1861 unsigned int current_read_size;
1862 struct cifs_sb_info *cifs_sb;
1863 struct cifsTconInfo *pTcon;
1865 char *current_offset;
1866 struct cifsFileInfo *open_file;
1867 int buf_type = CIFS_NO_BUFFER;
1870 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1872 if (file->private_data == NULL) {
1877 open_file = file->private_data;
1878 pTcon = tlink_tcon(open_file->tlink);
1880 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1881 cFYI(1, "attempting read on write only file instance");
1883 for (total_read = 0, current_offset = read_data;
1884 read_size > total_read;
1885 total_read += bytes_read, current_offset += bytes_read) {
1886 current_read_size = min_t(const int, read_size - total_read,
1888 /* For windows me and 9x we do not want to request more
1889 than it negotiated since it will refuse the read then */
1891 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1892 current_read_size = min_t(const int, current_read_size,
1893 pTcon->ses->server->maxBuf - 128);
1896 while (rc == -EAGAIN) {
1897 if ((open_file->invalidHandle) &&
1898 (!open_file->closePend)) {
1899 rc = cifs_reopen_file(open_file, true);
1903 rc = CIFSSMBRead(xid, pTcon,
1905 current_read_size, *poffset,
1906 &bytes_read, ¤t_offset,
1909 if (rc || (bytes_read == 0)) {
1917 cifs_stats_bytes_read(pTcon, total_read);
1918 *poffset += bytes_read;
1925 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1930 rc = cifs_revalidate_file(file);
1932 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1936 rc = generic_file_mmap(file, vma);
1942 static void cifs_copy_cache_pages(struct address_space *mapping,
1943 struct list_head *pages, int bytes_read, char *data)
1948 while (bytes_read > 0) {
1949 if (list_empty(pages))
1952 page = list_entry(pages->prev, struct page, lru);
1953 list_del(&page->lru);
1955 if (add_to_page_cache_lru(page, mapping, page->index,
1957 page_cache_release(page);
1958 cFYI(1, "Add page cache failed");
1959 data += PAGE_CACHE_SIZE;
1960 bytes_read -= PAGE_CACHE_SIZE;
1963 page_cache_release(page);
1965 target = kmap_atomic(page, KM_USER0);
1967 if (PAGE_CACHE_SIZE > bytes_read) {
1968 memcpy(target, data, bytes_read);
1969 /* zero the tail end of this partial page */
1970 memset(target + bytes_read, 0,
1971 PAGE_CACHE_SIZE - bytes_read);
1974 memcpy(target, data, PAGE_CACHE_SIZE);
1975 bytes_read -= PAGE_CACHE_SIZE;
1977 kunmap_atomic(target, KM_USER0);
1979 flush_dcache_page(page);
1980 SetPageUptodate(page);
1982 data += PAGE_CACHE_SIZE;
1984 /* add page to FS-Cache */
1985 cifs_readpage_to_fscache(mapping->host, page);
1990 static int cifs_readpages(struct file *file, struct address_space *mapping,
1991 struct list_head *page_list, unsigned num_pages)
1997 struct cifs_sb_info *cifs_sb;
1998 struct cifsTconInfo *pTcon;
1999 unsigned int bytes_read = 0;
2000 unsigned int read_size, i;
2001 char *smb_read_data = NULL;
2002 struct smb_com_read_rsp *pSMBr;
2003 struct cifsFileInfo *open_file;
2004 int buf_type = CIFS_NO_BUFFER;
2007 if (file->private_data == NULL) {
2012 open_file = file->private_data;
2013 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2014 pTcon = tlink_tcon(open_file->tlink);
2017 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2018 * immediately if the cookie is negative
2020 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2025 cFYI(DBG2, "rpages: num pages %d", num_pages);
2026 for (i = 0; i < num_pages; ) {
2027 unsigned contig_pages;
2028 struct page *tmp_page;
2029 unsigned long expected_index;
2031 if (list_empty(page_list))
2034 page = list_entry(page_list->prev, struct page, lru);
2035 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2037 /* count adjacent pages that we will read into */
2040 list_entry(page_list->prev, struct page, lru)->index;
2041 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2042 if (tmp_page->index == expected_index) {
2048 if (contig_pages + i > num_pages)
2049 contig_pages = num_pages - i;
2051 /* for reads over a certain size could initiate async
2054 read_size = contig_pages * PAGE_CACHE_SIZE;
2055 /* Read size needs to be in multiples of one page */
2056 read_size = min_t(const unsigned int, read_size,
2057 cifs_sb->rsize & PAGE_CACHE_MASK);
2058 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2059 read_size, contig_pages);
2061 while (rc == -EAGAIN) {
2062 if ((open_file->invalidHandle) &&
2063 (!open_file->closePend)) {
2064 rc = cifs_reopen_file(open_file, true);
2069 rc = CIFSSMBRead(xid, pTcon,
2072 &bytes_read, &smb_read_data,
2074 /* BB more RC checks ? */
2075 if (rc == -EAGAIN) {
2076 if (smb_read_data) {
2077 if (buf_type == CIFS_SMALL_BUFFER)
2078 cifs_small_buf_release(smb_read_data);
2079 else if (buf_type == CIFS_LARGE_BUFFER)
2080 cifs_buf_release(smb_read_data);
2081 smb_read_data = NULL;
2085 if ((rc < 0) || (smb_read_data == NULL)) {
2086 cFYI(1, "Read error in readpages: %d", rc);
2088 } else if (bytes_read > 0) {
2089 task_io_account_read(bytes_read);
2090 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2091 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2092 smb_read_data + 4 /* RFC1001 hdr */ +
2093 le16_to_cpu(pSMBr->DataOffset));
2095 i += bytes_read >> PAGE_CACHE_SHIFT;
2096 cifs_stats_bytes_read(pTcon, bytes_read);
2097 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2098 i++; /* account for partial page */
2100 /* server copy of file can have smaller size
2102 /* BB do we need to verify this common case ?
2103 this case is ok - if we are at server EOF
2104 we will hit it on next read */
2109 cFYI(1, "No bytes read (%d) at offset %lld . "
2110 "Cleaning remaining pages from readahead list",
2111 bytes_read, offset);
2112 /* BB turn off caching and do new lookup on
2113 file size at server? */
2116 if (smb_read_data) {
2117 if (buf_type == CIFS_SMALL_BUFFER)
2118 cifs_small_buf_release(smb_read_data);
2119 else if (buf_type == CIFS_LARGE_BUFFER)
2120 cifs_buf_release(smb_read_data);
2121 smb_read_data = NULL;
2126 /* need to free smb_read_data buf before exit */
2127 if (smb_read_data) {
2128 if (buf_type == CIFS_SMALL_BUFFER)
2129 cifs_small_buf_release(smb_read_data);
2130 else if (buf_type == CIFS_LARGE_BUFFER)
2131 cifs_buf_release(smb_read_data);
2132 smb_read_data = NULL;
2140 static int cifs_readpage_worker(struct file *file, struct page *page,
2146 /* Is the page cached? */
2147 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2151 page_cache_get(page);
2152 read_data = kmap(page);
2153 /* for reads over a certain size could initiate async read ahead */
2155 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2160 cFYI(1, "Bytes read %d", rc);
2162 file->f_path.dentry->d_inode->i_atime =
2163 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2165 if (PAGE_CACHE_SIZE > rc)
2166 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2168 flush_dcache_page(page);
2169 SetPageUptodate(page);
2171 /* send this page to the cache */
2172 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2178 page_cache_release(page);
2184 static int cifs_readpage(struct file *file, struct page *page)
2186 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2192 if (file->private_data == NULL) {
2198 cFYI(1, "readpage %p at offset %d 0x%x\n",
2199 page, (int)offset, (int)offset);
2201 rc = cifs_readpage_worker(file, page, &offset);
2209 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2211 struct cifsFileInfo *open_file;
2213 spin_lock(&cifs_file_list_lock);
2214 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2215 if (open_file->closePend)
2217 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2218 spin_unlock(&cifs_file_list_lock);
2222 spin_unlock(&cifs_file_list_lock);
2226 /* We do not want to update the file size from server for inodes
2227 open for write - to avoid races with writepage extending
2228 the file - in the future we could consider allowing
2229 refreshing the inode only on increases in the file size
2230 but this is tricky to do without racing with writebehind
2231 page caching in the current Linux kernel design */
2232 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2237 if (is_inode_writable(cifsInode)) {
2238 /* This inode is open for write at least once */
2239 struct cifs_sb_info *cifs_sb;
2241 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2242 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2243 /* since no page cache to corrupt on directio
2244 we can change size safely */
2248 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2256 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2257 loff_t pos, unsigned len, unsigned flags,
2258 struct page **pagep, void **fsdata)
2260 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2261 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2262 loff_t page_start = pos & PAGE_MASK;
2267 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2269 page = grab_cache_page_write_begin(mapping, index, flags);
2275 if (PageUptodate(page))
2279 * If we write a full page it will be up to date, no need to read from
2280 * the server. If the write is short, we'll end up doing a sync write
2283 if (len == PAGE_CACHE_SIZE)
2287 * optimize away the read when we have an oplock, and we're not
2288 * expecting to use any of the data we'd be reading in. That
2289 * is, when the page lies beyond the EOF, or straddles the EOF
2290 * and the write will cover all of the existing data.
2292 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2293 i_size = i_size_read(mapping->host);
2294 if (page_start >= i_size ||
2295 (offset == 0 && (pos + len) >= i_size)) {
2296 zero_user_segments(page, 0, offset,
2300 * PageChecked means that the parts of the page
2301 * to which we're not writing are considered up
2302 * to date. Once the data is copied to the
2303 * page, it can be set uptodate.
2305 SetPageChecked(page);
2310 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2312 * might as well read a page, it is fast enough. If we get
2313 * an error, we don't need to return it. cifs_write_end will
2314 * do a sync write instead since PG_uptodate isn't set.
2316 cifs_readpage_worker(file, page, &page_start);
2318 /* we could try using another file handle if there is one -
2319 but how would we lock it to prevent close of that handle
2320 racing with this read? In any case
2321 this will be written out by write_end so is fine */
2328 static int cifs_release_page(struct page *page, gfp_t gfp)
2330 if (PagePrivate(page))
2333 return cifs_fscache_release_page(page, gfp);
2336 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2338 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2341 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2344 void cifs_oplock_break(struct work_struct *work)
2346 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2348 struct inode *inode = cfile->dentry->d_inode;
2349 struct cifsInodeInfo *cinode = CIFS_I(inode);
2352 if (inode && S_ISREG(inode->i_mode)) {
2353 if (cinode->clientCanCacheRead)
2354 break_lease(inode, O_RDONLY);
2356 break_lease(inode, O_WRONLY);
2357 rc = filemap_fdatawrite(inode->i_mapping);
2358 if (cinode->clientCanCacheRead == 0) {
2359 waitrc = filemap_fdatawait(inode->i_mapping);
2360 invalidate_remote_inode(inode);
2365 cinode->write_behind_rc = rc;
2366 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2370 * releasing stale oplock after recent reconnect of smb session using
2371 * a now incorrect file handle is not a data integrity issue but do
2372 * not bother sending an oplock release if session to server still is
2373 * disconnected since oplock already released by the server
2375 if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2376 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2377 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2378 cFYI(1, "Oplock release rc = %d", rc);
2382 * We might have kicked in before is_valid_oplock_break()
2383 * finished grabbing reference for us. Make sure it's done by
2384 * waiting for GlobalSMSSeslock.
2386 spin_lock(&cifs_file_list_lock);
2387 spin_unlock(&cifs_file_list_lock);
2389 cifs_oplock_break_put(cfile);
2392 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2394 cifs_sb_active(cfile->dentry->d_sb);
2395 cifsFileInfo_get(cfile);
2398 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2400 cifsFileInfo_put(cfile);
2401 cifs_sb_deactive(cfile->dentry->d_sb);
2404 const struct address_space_operations cifs_addr_ops = {
2405 .readpage = cifs_readpage,
2406 .readpages = cifs_readpages,
2407 .writepage = cifs_writepage,
2408 .writepages = cifs_writepages,
2409 .write_begin = cifs_write_begin,
2410 .write_end = cifs_write_end,
2411 .set_page_dirty = __set_page_dirty_nobuffers,
2412 .releasepage = cifs_release_page,
2413 .invalidatepage = cifs_invalidate_page,
2414 /* .sync_page = cifs_sync_page, */
2419 * cifs_readpages requires the server to support a buffer large enough to
2420 * contain the header plus one complete page of data. Otherwise, we need
2421 * to leave cifs_readpages out of the address space operations.
2423 const struct address_space_operations cifs_addr_ops_smallbuf = {
2424 .readpage = cifs_readpage,
2425 .writepage = cifs_writepage,
2426 .writepages = cifs_writepages,
2427 .write_begin = cifs_write_begin,
2428 .write_end = cifs_write_end,
2429 .set_page_dirty = __set_page_dirty_nobuffers,
2430 .releasepage = cifs_release_page,
2431 .invalidatepage = cifs_invalidate_page,
2432 /* .sync_page = cifs_sync_page, */